🧪 test(l6): rewrite step-10 prompt to drop explicit consultant naming

Resolves GL !2923 (local #36).

The prior prompt told bro let cto weigh in — bypassing bro's consultant-classification judgment. The scenario is supposed to exercise tmb_agent-creator's description-match heuristic for picking the right consultant (architect / cto / pm / ceo).

New prompt: we're hitting scaling bottlenecks as traffic grows — should we break the auth service out of the monolith or keep it co-located? — a natural tech-strategy ask. tmb_agent-creator should auto-classify as cto via description match. End-to-end consultant-spawn validation preserved without the unrealistic consultant-naming shortcut.

Changes: 2 files (prompt.txt + README.md). Scorers unchanged (still applicable).

Same fix pattern as #2904 (closed) Phase B step-14 commit 967d448.

Verification: bash tests/run-all.sh exit 0 on this branch (rebased onto dev with !2922 + !2924 merged). All L1+L2+L3+L4 green.

pr-reviewer verdict: PASS (validation_attempts id=20).